NAME
   AnyEvent - the DBI of event loop programming

   EV, Event, Glib, Tk, UV, Perl, Event::Lib, Irssi, rxvt-unicode,
   IO::Async, Qt, FLTK and POE are various supported event
   loops/environments.

SYNOPSIS
      use AnyEvent;

      # if you prefer function calls, look at the AE manpage for
      # an alternative API.

      # file handle or descriptor readable
      my $w = AnyEvent->io (fh => $fh, poll => "r", cb => sub { ...  });

      # one-shot or repeating timers
      my $w = AnyEvent->timer (after => $seconds, cb => sub { ...  });
      my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ...);

      print AnyEvent->now;  # prints current event loop time
      print AnyEvent->time; # think Time::HiRes::time or simply CORE::time.

      # POSIX signal
      my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... });

      # child process exit
      my $w = AnyEvent->child (pid => $pid, cb => sub {
         my ($pid, $status) = @_;
         ...
      });

      # called when event loop idle (if applicable)
      my $w = AnyEvent->idle (cb => sub { ... });

      my $w = AnyEvent->condvar; # stores whether a condition was flagged
      $w->send; # wake up current and all future recv's
      $w->recv; # enters "main loop" till $condvar gets ->send
      # use a condvar in callback mode:
      $w->cb (sub { $_[0]->recv });

INTRODUCTION/TUTORIAL
   This manpage is mainly a reference manual. If you are interested in a
   tutorial or some gentle introduction, have a look at the AnyEvent::Intro
   manpage.

SUPPORT
   An FAQ document is available as AnyEvent::FAQ.

   There also is a mailinglist for discussing all things AnyEvent, and an
   IRC channel, too.

   See the AnyEvent project page at the Schmorpforge Ta-Sa Software
   Repository, at <http://anyevent.schmorp.de>, for more info.

WHY YOU SHOULD USE THIS MODULE (OR NOT)
   Glib, POE, IO::Async, Event... CPAN offers event models by the dozen
   nowadays. So what is different about AnyEvent?

   Executive Summary: AnyEvent is *compatible*, AnyEvent is *free of
   policy* and AnyEvent is *small and efficient*.

   First and foremost, *AnyEvent is not an event model* itself, it only
   interfaces to whatever event model the main program happens to use, in a
   pragmatic way. For event models and certain classes of immortals alike,
   the statement "there can only be one" is a bitter reality: In general,
   only one event loop can be active at the same time in a process.
   AnyEvent cannot change this, but it can hide the differences between
   those event loops.

   The goal of AnyEvent is to offer module authors the ability to do event
   programming (waiting for I/O or timer events) without subscribing to a
   religion, a way of living, and most importantly: without forcing your
   module users into the same thing by forcing them to use the same event
   model you use.

   For modules like POE or IO::Async (which is a total misnomer as it is
   actually doing all I/O *synchronously*...), using them in your module is
   like joining a cult: After you join, you are dependent on them and you
   cannot use anything else, as they are simply incompatible to everything
   that isn't them. What's worse, all the potential users of your module
   are *also* forced to use the same event loop you use.

   AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works
   fine. AnyEvent + Tk works fine etc. etc. but none of these work together
   with the rest: POE + EV? No go. Tk + Event? No go. Again: if your module
   uses one of those, every user of your module has to use it, too. But if
   your module uses AnyEvent, it works transparently with all event models
   it supports (including stuff like IO::Async, as long as those use one of
   the supported event loops. It is easy to add new event loops to
   AnyEvent, too, so it is future-proof).

   In addition to being free of having to use *the one and only true event
   model*, AnyEvent also is free of bloat and policy: with POE or similar
   modules, you get an enormous amount of code and strict rules you have to
   follow. AnyEvent, on the other hand, is lean and to the point, by only
   offering the functionality that is necessary, in as thin as a wrapper as
   technically possible.

   Of course, AnyEvent comes with a big (and fully optional!) toolbox of
   useful functionality, such as an asynchronous DNS resolver, 100%
   non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms
   such as Windows) and lots of real-world knowledge and workarounds for
   platform bugs and differences.

   Now, if you *do want* lots of policy (this can arguably be somewhat
   useful) and you want to force your users to use the one and only event
   model, you should *not* use this module.

DESCRIPTION
   AnyEvent provides a uniform interface to various event loops. This
   allows module authors to use event loop functionality without forcing
   module users to use a specific event loop implementation (since more
   than one event loop cannot coexist peacefully).

   The interface itself is vaguely similar, but not identical to the Event
   module.

   During the first call of any watcher-creation method, the module tries
   to detect the currently loaded event loop by probing whether one of the
   following modules is already loaded: EV, AnyEvent::Loop, Event, Glib,
   Tk, Event::Lib, Qt, POE. The first one found is used. If none are
   detected, the module tries to load the first four modules in the order
   given; but note that if EV is not available, the pure-perl
   AnyEvent::Loop should always work, so the other two are not normally
   tried.

   Because AnyEvent first checks for modules that are already loaded,
   loading an event model explicitly before first using AnyEvent will
   likely make that model the default. For example:

      use Tk;
      use AnyEvent;

      # .. AnyEvent will likely default to Tk

   The *likely* means that, if any module loads another event model and
   starts using it, all bets are off - this case should be very rare
   though, as very few modules hardcode event loops without announcing this
   very loudly.

   The pure-perl implementation of AnyEvent is called "AnyEvent::Loop".
   Like other event modules you can load it explicitly and enjoy the high
   availability of that event loop :)

WATCHERS
   AnyEvent has the central concept of a *watcher*, which is an object that
   stores relevant data for each kind of event you are waiting for, such as
   the callback to call, the file handle to watch, etc.

   These watchers are normal Perl objects with normal Perl lifetime. After
   creating a watcher it will immediately "watch" for events and invoke the
   callback when the event occurs (of course, only when the event model is
   in control).

   Note that callbacks must not permanently change global variables
   potentially in use by the event loop (such as $_ or $[) and that
   callbacks must not "die". The former is good programming practice in
   Perl and the latter stems from the fact that exception handling differs
   widely between event loops.

   To disable a watcher you have to destroy it (e.g. by setting the
   variable you store it in to "undef" or otherwise deleting all references
   to it).

   All watchers are created by calling a method on the "AnyEvent" class.

   Many watchers either are used with "recursion" (repeating timers for
   example), or need to refer to their watcher object in other ways.

   One way to achieve that is this pattern:

      my $w; $w = AnyEvent->type (arg => value ..., cb => sub {
         # you can use $w here, for example to undef it
         undef $w;
      });

   Note that "my $w; $w =" combination. This is necessary because in Perl,
   my variables are only visible after the statement in which they are
   declared.

 I/O WATCHERS
      $w = AnyEvent->io (
         fh   => <filehandle_or_fileno>,
         poll => <"r" or "w">,
         cb   => <callback>,
      );

   You can create an I/O watcher by calling the "AnyEvent->io" method with
   the following mandatory key-value pairs as arguments:

   "fh" is the Perl *file handle* (or a naked file descriptor) to watch for
   events (AnyEvent might or might not keep a reference to this file
   handle). Note that only file handles pointing to things for which
   non-blocking operation makes sense are allowed. This includes sockets,
   most character devices, pipes, fifos and so on, but not for example
   files or block devices.

   "poll" must be a string that is either "r" or "w", which creates a
   watcher waiting for "r"eadable or "w"ritable events, respectively.

   "cb" is the callback to invoke each time the file handle becomes ready.

   Although the callback might get passed parameters, their value and
   presence is undefined and you cannot rely on them. Portable AnyEvent
   callbacks cannot use arguments passed to I/O watcher callbacks.

   The I/O watcher might use the underlying file descriptor or a copy of
   it. You must not close a file handle as long as any watcher is active on
   the underlying file descriptor.

   Some event loops issue spurious readiness notifications, so you should
   always use non-blocking calls when reading/writing from/to your file
   handles.

   Example: wait for readability of STDIN, then read a line and disable the
   watcher.

      my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
         chomp (my $input = <STDIN>);
         warn "read: $input\n";
         undef $w;
      });

 TIME WATCHERS
      $w = AnyEvent->timer (after => <seconds>, cb => <callback>);

      $w = AnyEvent->timer (
         after    => <fractional_seconds>,
         interval => <fractional_seconds>,
         cb       => <callback>,
      );

   You can create a time watcher by calling the "AnyEvent->timer" method
   with the following mandatory arguments:

   "after" specifies after how many seconds (fractional values are
   supported) the callback should be invoked. "cb" is the callback to
   invoke in that case.

   Although the callback might get passed parameters, their value and
   presence is undefined and you cannot rely on them. Portable AnyEvent
   callbacks cannot use arguments passed to time watcher callbacks.

   The callback will normally be invoked only once. If you specify another
   parameter, "interval", as a strictly positive number (> 0), then the
   callback will be invoked regularly at that interval (in fractional
   seconds) after the first invocation. If "interval" is specified with a
   false value, then it is treated as if it were not specified at all.

   The callback will be rescheduled before invoking the callback, but no
   attempt is made to avoid timer drift in most backends, so the interval
   is only approximate.

   Example: fire an event after 7.7 seconds.

      my $w = AnyEvent->timer (after => 7.7, cb => sub {
         warn "timeout\n";
      });

      # to cancel the timer:
      undef $w;

   Example 2: fire an event after 0.5 seconds, then roughly every second.

      my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub {
         warn "timeout\n";
      });

  TIMING ISSUES
   There are two ways to handle timers: based on real time (relative, "fire
   in 10 seconds") and based on wallclock time (absolute, "fire at 12
   o'clock").

   While most event loops expect timers to specified in a relative way,
   they use absolute time internally. This makes a difference when your
   clock "jumps", for example, when ntp decides to set your clock backwards
   from the wrong date of 2014-01-01 to 2008-01-01, a watcher that is
   supposed to fire "after a second" might actually take six years to
   finally fire.

   AnyEvent cannot compensate for this. The only event loop that is
   conscious of these issues is EV, which offers both relative (ev_timer,
   based on true relative time) and absolute (ev_periodic, based on
   wallclock time) timers.

   AnyEvent always prefers relative timers, if available, matching the
   AnyEvent API.

   AnyEvent has two additional methods that return the "current time":

   AnyEvent->time
       This returns the "current wallclock time" as a fractional number of
       seconds since the Epoch (the same thing as "time" or
       "Time::HiRes::time" return, and the result is guaranteed to be
       compatible with those).

       It progresses independently of any event loop processing, i.e. each
       call will check the system clock, which usually gets updated
       frequently.

   AnyEvent->now
       This also returns the "current wallclock time", but unlike "time",
       above, this value might change only once per event loop iteration,
       depending on the event loop (most return the same time as "time",
       above). This is the time that AnyEvent's timers get scheduled
       against.

       *In almost all cases (in all cases if you don't care), this is the
       function to call when you want to know the current time.*

       This function is also often faster then "AnyEvent->time", and thus
       the preferred method if you want some timestamp (for example,
       AnyEvent::Handle uses this to update its activity timeouts).

       The rest of this section is only of relevance if you try to be very
       exact with your timing; you can skip it without a bad conscience.

       For a practical example of when these times differ, consider
       Event::Lib and EV and the following set-up:

       The event loop is running and has just invoked one of your callbacks
       at time=500 (assume no other callbacks delay processing). In your
       callback, you wait a second by executing "sleep 1" (blocking the
       process for a second) and then (at time=501) you create a relative
       timer that fires after three seconds.

       With Event::Lib, "AnyEvent->time" and "AnyEvent->now" will both
       return 501, because that is the current time, and the timer will be
       scheduled to fire at time=504 (501 + 3).

       With EV, "AnyEvent->time" returns 501 (as that is the current time),
       but "AnyEvent->now" returns 500, as that is the time the last event
       processing phase started. With EV, your timer gets scheduled to run
       at time=503 (500 + 3).

       In one sense, Event::Lib is more exact, as it uses the current time
       regardless of any delays introduced by event processing. However,
       most callbacks do not expect large delays in processing, so this
       causes a higher drift (and a lot more system calls to get the
       current time).

       In another sense, EV is more exact, as your timer will be scheduled
       at the same time, regardless of how long event processing actually
       took.

       In either case, if you care (and in most cases, you don't), then you
       can get whatever behaviour you want with any event loop, by taking
       the difference between "AnyEvent->time" and "AnyEvent->now" into
       account.

   AnyEvent->now_update
       Some event loops (such as EV or AnyEvent::Loop) cache the current
       time for each loop iteration (see the discussion of AnyEvent->now,
       above).

       When a callback runs for a long time (or when the process sleeps),
       then this "current" time will differ substantially from the real
       time, which might affect timers and time-outs.

       When this is the case, you can call this method, which will update
       the event loop's idea of "current time".

       A typical example would be a script in a web server (e.g.
       "mod_perl") - when mod_perl executes the script, then the event loop
       will have the wrong idea about the "current time" (being potentially
       far in the past, when the script ran the last time). In that case
       you should arrange a call to "AnyEvent->now_update" each time the
       web server process wakes up again (e.g. at the start of your script,
       or in a handler).

       Note that updating the time *might* cause some events to be handled.

 SIGNAL WATCHERS
      $w = AnyEvent->signal (signal => <uppercase_signal_name>, cb => <callback>);

   You can watch for signals using a signal watcher, "signal" is the signal
   *name* in uppercase and without any "SIG" prefix, "cb" is the Perl
   callback to be invoked whenever a signal occurs.

   Although the callback might get passed parameters, their value and
   presence is undefined and you cannot rely on them. Portable AnyEvent
   callbacks cannot use arguments passed to signal watcher callbacks.

   Multiple signal occurrences can be clumped together into one callback
   invocation, and callback invocation will be synchronous. Synchronous
   means that it might take a while until the signal gets handled by the
   process, but it is guaranteed not to interrupt any other callbacks.

   The main advantage of using these watchers is that you can share a
   signal between multiple watchers, and AnyEvent will ensure that signals
   will not interrupt your program at bad times.

   This watcher might use %SIG (depending on the event loop used), so
   programs overwriting those signals directly will likely not work
   correctly.

   Example: exit on SIGINT

      my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 });

  Restart Behaviour
   While restart behaviour is up to the event loop implementation, most
   will not restart syscalls (that includes Async::Interrupt and AnyEvent's
   pure perl implementation).

  Safe/Unsafe Signals
   Perl signals can be either "safe" (synchronous to opcode handling) or
   "unsafe" (asynchronous) - the former might delay signal delivery
   indefinitely, the latter might corrupt your memory.

   AnyEvent signal handlers are, in addition, synchronous to the event
   loop, i.e. they will not interrupt your running perl program but will
   only be called as part of the normal event handling (just like timer,
   I/O etc. callbacks, too).

  Signal Races, Delays and Workarounds
   Many event loops (e.g. Glib, Tk, Qt, IO::Async) do not support attaching
   callbacks to signals in a generic way, which is a pity, as you cannot do
   race-free signal handling in perl, requiring C libraries for this.
   AnyEvent will try to do its best, which means in some cases, signals
   will be delayed. The maximum time a signal might be delayed is 10
   seconds by default, but can be overriden via
   $ENV{PERL_ANYEVENT_MAX_SIGNAL_LATENCY} or $AnyEvent::MAX_SIGNAL_LATENCY
   - see the "ENVIRONMENT VARIABLES" section for details.

   All these problems can be avoided by installing the optional
   Async::Interrupt module, which works with most event loops. It will not
   work with inherently broken event loops such as Event or Event::Lib (and
   not with POE currently). For those, you just have to suffer the delays.

 CHILD PROCESS WATCHERS
      $w = AnyEvent->child (pid => <process id>, cb => <callback>);

   You can also watch for a child process exit and catch its exit status.

   The child process is specified by the "pid" argument (on some backends,
   using 0 watches for any child process exit, on others this will croak).
   The watcher will be triggered only when the child process has finished
   and an exit status is available, not on any trace events
   (stopped/continued).

   The callback will be called with the pid and exit status (as returned by
   waitpid), so unlike other watcher types, you *can* rely on child watcher
   callback arguments.

   This watcher type works by installing a signal handler for "SIGCHLD",
   and since it cannot be shared, nothing else should use SIGCHLD or reap
   random child processes (waiting for specific child processes, e.g.
   inside "system", is just fine).

   There is a slight catch to child watchers, however: you usually start
   them *after* the child process was created, and this means the process
   could have exited already (and no SIGCHLD will be sent anymore).

   Not all event models handle this correctly (neither POE nor IO::Async
   do, see their AnyEvent::Impl manpages for details), but even for event
   models that *do* handle this correctly, they usually need to be loaded
   before the process exits (i.e. before you fork in the first place).
   AnyEvent's pure perl event loop handles all cases correctly regardless
   of when you start the watcher.

   This means you cannot create a child watcher as the very first thing in
   an AnyEvent program, you *have* to create at least one watcher before
   you "fork" the child (alternatively, you can call "AnyEvent::detect").

   As most event loops do not support waiting for child events, they will
   be emulated by AnyEvent in most cases, in which case the latency and
   race problems mentioned in the description of signal watchers apply.

   Example: fork a process and wait for it

      my $done = AnyEvent->condvar;

      # this forks and immediately calls exit in the child. this
      # normally has all sorts of bad consequences for your parent,
      # so take this as an example only. always fork and exec,
      # or call POSIX::_exit, in real code.
      my $pid = fork or exit 5;

      my $w = AnyEvent->child (
         pid => $pid,
         cb  => sub {
            my ($pid, $status) = @_;
            warn "pid $pid exited with status $status";
            $done->send;
         },
      );

      # do something else, then wait for process exit
      $done->recv;

 IDLE WATCHERS
      $w = AnyEvent->idle (cb => <callback>);

   This will repeatedly invoke the callback after the process becomes idle,
   until either the watcher is destroyed or new events have been detected.

   Idle watchers are useful when there is a need to do something, but it is
   not so important (or wise) to do it instantly. The callback will be
   invoked only when there is "nothing better to do", which is usually
   defined as "all outstanding events have been handled and no new events
   have been detected". That means that idle watchers ideally get invoked
   when the event loop has just polled for new events but none have been
   detected. Instead of blocking to wait for more events, the idle watchers
   will be invoked.

   Unfortunately, most event loops do not really support idle watchers
   (only EV, Event and Glib do it in a usable fashion) - for the rest,
   AnyEvent will simply call the callback "from time to time".

   Example: read lines from STDIN, but only process them when the program
   is otherwise idle:

      my @lines; # read data
      my $idle_w;
      my $io_w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub {
         push @lines, scalar <STDIN>;

         # start an idle watcher, if not already done
         $idle_w ||= AnyEvent->idle (cb => sub {
            # handle only one line, when there are lines left
            if (my $line = shift @lines) {
               print "handled when idle: $line";
            } else {
               # otherwise disable the idle watcher again
               undef $idle_w;
            }
         });
      });

 CONDITION VARIABLES
      $cv = AnyEvent->condvar;

      $cv->send (<list>);
      my @res = $cv->recv;

   If you are familiar with some event loops you will know that all of them
   require you to run some blocking "loop", "run" or similar function that
   will actively watch for new events and call your callbacks.

   AnyEvent is slightly different: it expects somebody else to run the
   event loop and will only block when necessary (usually when told by the
   user).

   The tool to do that is called a "condition variable", so called because
   they represent a condition that must become true.

   Now is probably a good time to look at the examples further below.

   Condition variables can be created by calling the "AnyEvent->condvar"
   method, usually without arguments. The only argument pair allowed is
   "cb", which specifies a callback to be called when the condition
   variable becomes true, with the condition variable as the first argument
   (but not the results).

   After creation, the condition variable is "false" until it becomes
   "true" by calling the "send" method (or calling the condition variable
   as if it were a callback, read about the caveats in the description for
   the "->send" method).

   Since condition variables are the most complex part of the AnyEvent API,
   here are some different mental models of what they are - pick the ones
   you can connect to:

   *   Condition variables are like callbacks - you can call them (and pass
       them instead of callbacks). Unlike callbacks however, you can also
       wait for them to be called.

   *   Condition variables are signals - one side can emit or send them,
       the other side can wait for them, or install a handler that is
       called when the signal fires.

   *   Condition variables are like "Merge Points" - points in your program
       where you merge multiple independent results/control flows into one.

   *   Condition variables represent a transaction - functions that start
       some kind of transaction can return them, leaving the caller the
       choice between waiting in a blocking fashion, or setting a callback.

   *   Condition variables represent future values, or promises to deliver
       some result, long before the result is available.

   Condition variables are very useful to signal that something has
   finished, for example, if you write a module that does asynchronous http
   requests, then a condition variable would be the ideal candidate to
   signal the availability of results. The user can either act when the
   callback is called or can synchronously "->recv" for the results.

   You can also use them to simulate traditional event loops - for example,
   you can block your main program until an event occurs - for example, you
   could "->recv" in your main program until the user clicks the Quit
   button of your app, which would "->send" the "quit" event.

   Note that condition variables recurse into the event loop - if you have
   two pieces of code that call "->recv" in a round-robin fashion, you
   lose. Therefore, condition variables are good to export to your caller,
   but you should avoid making a blocking wait yourself, at least in
   callbacks, as this asks for trouble.

   Condition variables are represented by hash refs in perl, and the keys
   used by AnyEvent itself are all named "_ae_XXX" to make subclassing easy
   (it is often useful to build your own transaction class on top of
   AnyEvent). To subclass, use "AnyEvent::CondVar" as base class and call
   its "new" method in your own "new" method.

   There are two "sides" to a condition variable - the "producer side"
   which eventually calls "-> send", and the "consumer side", which waits
   for the send to occur.

   Example: wait for a timer.

      # condition: "wait till the timer is fired"
      my $timer_fired = AnyEvent->condvar;

      # create the timer - we could wait for, say
      # a handle becomign ready, or even an
      # AnyEvent::HTTP request to finish, but
      # in this case, we simply use a timer:
      my $w = AnyEvent->timer (
         after => 1,
         cb    => sub { $timer_fired->send },
      );

      # this "blocks" (while handling events) till the callback
      # calls ->send
      $timer_fired->recv;

   Example: wait for a timer, but take advantage of the fact that condition
   variables are also callable directly.

      my $done = AnyEvent->condvar;
      my $delay = AnyEvent->timer (after => 5, cb => $done);
      $done->recv;

   Example: Imagine an API that returns a condvar and doesn't support
   callbacks. This is how you make a synchronous call, for example from the
   main program:

      use AnyEvent::CouchDB;

      ...

      my @info = $couchdb->info->recv;

   And this is how you would just set a callback to be called whenever the
   results are available:

      $couchdb->info->cb (sub {
         my @info = $_[0]->recv;
      });

  METHODS FOR PRODUCERS
   These methods should only be used by the producing side, i.e. the
   code/module that eventually sends the signal. Note that it is also the
   producer side which creates the condvar in most cases, but it isn't
   uncommon for the consumer to create it as well.

   $cv->send (...)
       Flag the condition as ready - a running "->recv" and all further
       calls to "recv" will (eventually) return after this method has been
       called. If nobody is waiting the send will be remembered.

       If a callback has been set on the condition variable, it is called
       immediately from within send.

       Any arguments passed to the "send" call will be returned by all
       future "->recv" calls.

       Condition variables are overloaded so one can call them directly (as
       if they were a code reference). Calling them directly is the same as
       calling "send".

   $cv->croak ($error)
       Similar to send, but causes all calls to "->recv" to invoke
       "Carp::croak" with the given error message/object/scalar.

       This can be used to signal any errors to the condition variable
       user/consumer. Doing it this way instead of calling "croak" directly
       delays the error detection, but has the overwhelming advantage that
       it diagnoses the error at the place where the result is expected,
       and not deep in some event callback with no connection to the actual
       code causing the problem.

   $cv->begin ([group callback])
   $cv->end
       These two methods can be used to combine many transactions/events
       into one. For example, a function that pings many hosts in parallel
       might want to use a condition variable for the whole process.

       Every call to "->begin" will increment a counter, and every call to
       "->end" will decrement it. If the counter reaches 0 in "->end", the
       (last) callback passed to "begin" will be executed, passing the
       condvar as first argument. That callback is *supposed* to call
       "->send", but that is not required. If no group callback was set,
       "send" will be called without any arguments.

       You can think of "$cv->send" giving you an OR condition (one call
       sends), while "$cv->begin" and "$cv->end" giving you an AND
       condition (all "begin" calls must be "end"'ed before the condvar
       sends).

       Let's start with a simple example: you have two I/O watchers (for
       example, STDOUT and STDERR for a program), and you want to wait for
       both streams to close before activating a condvar:

          my $cv = AnyEvent->condvar;

          $cv->begin; # first watcher
          my $w1 = AnyEvent->io (fh => $fh1, cb => sub {
             defined sysread $fh1, my $buf, 4096
                or $cv->end;
          });

          $cv->begin; # second watcher
          my $w2 = AnyEvent->io (fh => $fh2, cb => sub {
             defined sysread $fh2, my $buf, 4096
                or $cv->end;
          });

          $cv->recv;

       This works because for every event source (EOF on file handle),
       there is one call to "begin", so the condvar waits for all calls to
       "end" before sending.

       The ping example mentioned above is slightly more complicated, as
       the there are results to be passed back, and the number of tasks
       that are begun can potentially be zero:

          my $cv = AnyEvent->condvar;

          my %result;
          $cv->begin (sub { shift->send (\%result) });

          for my $host (@list_of_hosts) {
             $cv->begin;
             ping_host_then_call_callback $host, sub {
                $result{$host} = ...;
                $cv->end;
             };
          }

          $cv->end;

          ...

          my $results = $cv->recv;

       This code fragment supposedly pings a number of hosts and calls
       "send" after results for all then have have been gathered - in any
       order. To achieve this, the code issues a call to "begin" when it
       starts each ping request and calls "end" when it has received some
       result for it. Since "begin" and "end" only maintain a counter, the
       order in which results arrive is not relevant.

       There is an additional bracketing call to "begin" and "end" outside
       the loop, which serves two important purposes: first, it sets the
       callback to be called once the counter reaches 0, and second, it
       ensures that "send" is called even when "no" hosts are being pinged
       (the loop doesn't execute once).

       This is the general pattern when you "fan out" into multiple (but
       potentially zero) subrequests: use an outer "begin"/"end" pair to
       set the callback and ensure "end" is called at least once, and then,
       for each subrequest you start, call "begin" and for each subrequest
       you finish, call "end".

  METHODS FOR CONSUMERS
   These methods should only be used by the consuming side, i.e. the code
   awaits the condition.

   $cv->recv
       Wait (blocking if necessary) until the "->send" or "->croak" methods
       have been called on $cv, while servicing other watchers normally.

       You can only wait once on a condition - additional calls are valid
       but will return immediately.

       If an error condition has been set by calling "->croak", then this
       function will call "croak".

       In list context, all parameters passed to "send" will be returned,
       in scalar context only the first one will be returned.

       Note that doing a blocking wait in a callback is not supported by
       any event loop, that is, recursive invocation of a blocking "->recv"
       is not allowed and the "recv" call will "croak" if such a condition
       is detected. This requirement can be dropped by relying on
       Coro::AnyEvent , which allows you to do a blocking "->recv" from any
       thread that doesn't run the event loop itself. Coro::AnyEvent is
       loaded automatically when Coro is used with AnyEvent, so code does
       not need to do anything special to take advantage of that: any code
       that would normally block your program because it calls "recv", be
       executed in an "async" thread instead without blocking other
       threads.

       Not all event models support a blocking wait - some die in that case
       (programs might want to do that to stay interactive), so *if you are
       using this from a module, never require a blocking wait*. Instead,
       let the caller decide whether the call will block or not (for
       example, by coupling condition variables with some kind of request
       results and supporting callbacks so the caller knows that getting
       the result will not block, while still supporting blocking waits if
       the caller so desires).

       You can ensure that "->recv" never blocks by setting a callback and
       only calling "->recv" from within that callback (or at a later
       time). This will work even when the event loop does not support
       blocking waits otherwise.

   $bool = $cv->ready
       Returns true when the condition is "true", i.e. whether "send" or
       "croak" have been called.

   $cb = $cv->cb ($cb->($cv))
       This is a mutator function that returns the callback set (or "undef"
       if not) and optionally replaces it before doing so.

       The callback will be called when the condition becomes "true", i.e.
       when "send" or "croak" are called, with the only argument being the
       condition variable itself. If the condition is already true, the
       callback is called immediately when it is set. Calling "recv" inside
       the callback or at any later time is guaranteed not to block.

       Additionally, when the callback is invoked, it is also removed from
       the condvar (reset to "undef"), so the condvar does not keep a
       reference to the callback after invocation.

SUPPORTED EVENT LOOPS/BACKENDS
   The following backend classes are part of the AnyEvent distribution
   (every class has its own manpage):

   Backends that are autoprobed when no other event loop can be found.
       EV is the preferred backend when no other event loop seems to be in
       use. If EV is not installed, then AnyEvent will fall back to its own
       pure-perl implementation, which is available everywhere as it comes
       with AnyEvent itself.

          AnyEvent::Impl::EV        based on EV (interface to libev, best choice).
          AnyEvent::Impl::Perl      pure-perl AnyEvent::Loop, fast and portable.

   Backends that are transparently being picked up when they are used.
       These will be used if they are already loaded when the first watcher
       is created, in which case it is assumed that the application is
       using them. This means that AnyEvent will automatically pick the
       right backend when the main program loads an event module before
       anything starts to create watchers. Nothing special needs to be done
       by the main program.

          AnyEvent::Impl::Event     based on Event, very stable, few glitches.
          AnyEvent::Impl::Glib      based on Glib, slow but very stable.
          AnyEvent::Impl::Tk        based on Tk, very broken.
          AnyEvent::Impl::UV        based on UV, innovated square wheels.
          AnyEvent::Impl::EventLib  based on Event::Lib, leaks memory and worse.
          AnyEvent::Impl::POE       based on POE, very slow, some limitations.
          AnyEvent::Impl::Irssi     used when running within irssi.
          AnyEvent::Impl::IOAsync   based on IO::Async.
          AnyEvent::Impl::Cocoa     based on Cocoa::EventLoop.
          AnyEvent::Impl::FLTK      based on FLTK (fltk 2 binding).

   Backends with special needs.
       Qt requires the Qt::Application to be instantiated first, but will
       otherwise be picked up automatically. As long as the main program
       instantiates the application before any AnyEvent watchers are
       created, everything should just work.

          AnyEvent::Impl::Qt        based on Qt.

   Event loops that are indirectly supported via other backends.
       Some event loops can be supported via other modules:

       There is no direct support for WxWidgets (Wx) or Prima.

       WxWidgets has no support for watching file handles. However, you can
       use WxWidgets through the POE adaptor, as POE has a Wx backend that
       simply polls 20 times per second, which was considered to be too
       horrible to even consider for AnyEvent.

       Prima is not supported as nobody seems to be using it, but it has a
       POE backend, so it can be supported through POE.

       AnyEvent knows about both Prima and Wx, however, and will try to
       load POE when detecting them, in the hope that POE will pick them
       up, in which case everything will be automatic.

   Known event loops outside the AnyEvent distribution
       The following event loops or programs support AnyEvent by providing
       their own AnyEvent backend. They will be picked up automatically.

          urxvt::anyevent           available to rxvt-unicode extensions

GLOBAL VARIABLES AND FUNCTIONS
   These are not normally required to use AnyEvent, but can be useful to
   write AnyEvent extension modules.

   $AnyEvent::MODEL
       Contains "undef" until the first watcher is being created, before
       the backend has been autodetected.

       Afterwards it contains the event model that is being used, which is
       the name of the Perl class implementing the model. This class is
       usually one of the "AnyEvent::Impl::xxx" modules, but can be any
       other class in the case AnyEvent has been extended at runtime (e.g.
       in *rxvt-unicode* it will be "urxvt::anyevent").

   AnyEvent::detect
       Returns $AnyEvent::MODEL, forcing autodetection of the event model
       if necessary. You should only call this function right before you
       would have created an AnyEvent watcher anyway, that is, as late as
       possible at runtime, and not e.g. during initialisation of your
       module.

       The effect of calling this function is as if a watcher had been
       created (specifically, actions that happen "when the first watcher
       is created" happen when calling detetc as well).

       If you need to do some initialisation before AnyEvent watchers are
       created, use "post_detect".

   $guard = AnyEvent::post_detect { BLOCK }
       Arranges for the code block to be executed as soon as the event
       model is autodetected (or immediately if that has already happened).

       The block will be executed *after* the actual backend has been
       detected ($AnyEvent::MODEL is set), so it is possible to do some
       initialisation only when AnyEvent is actually initialised - see the
       sources of AnyEvent::AIO to see how this is used.

       The most common usage is to create some global watchers, without
       forcing event module detection too early. For example, AnyEvent::AIO
       creates and installs the global IO::AIO watcher in a "post_detect"
       block to avoid autodetecting the event module at load time.

       If called in scalar or list context, then it creates and returns an
       object that automatically removes the callback again when it is
       destroyed (or "undef" when the hook was immediately executed). See
       AnyEvent::AIO for a case where this is useful.

       Example: Create a watcher for the IO::AIO module and store it in
       $WATCHER, but do so only do so after the event loop is initialised.

          our WATCHER;

          my $guard = AnyEvent::post_detect {
             $WATCHER = AnyEvent->io (fh => IO::AIO::poll_fileno, poll => 'r', cb => \&IO::AIO::poll_cb);
          };

          # the ||= is important in case post_detect immediately runs the block,
          # as to not clobber the newly-created watcher. assigning both watcher and
          # post_detect guard to the same variable has the advantage of users being
          # able to just C<undef $WATCHER> if the watcher causes them grief.

          $WATCHER ||= $guard;

   @AnyEvent::post_detect
       This is a lower level interface then "AnyEvent::post_detect" (the
       function). This variable is mainly useful for modules that can do
       something useful when AnyEvent is used and thus want to know when it
       is initialised, but do not need to even load it by default. This
       array provides the means to hook into AnyEvent passively, without
       loading it.

       Here is how it works: If there are any code references in this array
       (you can "push" to it before or after loading AnyEvent), then they
       will be called directly after the event loop has been chosen.

       You should check $AnyEvent::MODEL before adding to this array,
       though: if it is defined then the event loop has already been
       detected, and the array will be ignored.

       Best use "AnyEvent::post_detect { BLOCK }" when your application
       allows it, as it takes care of these details.

       Example: To load Coro::AnyEvent whenever Coro and AnyEvent are used
       together, you could put this into Coro (this is the actual code used
       by Coro to accomplish this):

          if (defined $AnyEvent::MODEL) {
             # AnyEvent already initialised, so load Coro::AnyEvent
             require Coro::AnyEvent;
          } else {
             # AnyEvent not yet initialised, so make sure to load Coro::AnyEvent
             # as soon as it is
             push @AnyEvent::post_detect, sub { require Coro::AnyEvent };
          }

   AnyEvent::postpone { BLOCK }
       Arranges for the block to be executed as soon as possible, but not
       before the call itself returns. In practise, the block will be
       executed just before the event loop polls for new events, or shortly
       afterwards.

       This function never returns anything (to make the "return postpone {
       ... }" idiom more useful.

       To understand the usefulness of this function, consider a function
       that asynchronously does something for you and returns some
       transaction object or guard to let you cancel the operation. For
       example, "AnyEvent::Socket::tcp_connect":

          # start a connection attempt unless one is active
          $self->{connect_guard} ||= AnyEvent::Socket::tcp_connect "www.example.net", 80, sub {
             delete $self->{connect_guard};
             ...
          };

       Imagine that this function could instantly call the callback, for
       example, because it detects an obvious error such as a negative port
       number. Invoking the callback before the function returns causes
       problems however: the callback will be called and will try to delete
       the guard object. But since the function hasn't returned yet, there
       is nothing to delete. When the function eventually returns it will
       assign the guard object to "$self->{connect_guard}", where it will
       likely never be deleted, so the program thinks it is still trying to
       connect.

       This is where "AnyEvent::postpone" should be used. Instead of
       calling the callback directly on error:

          $cb->(undef), return # signal error to callback, BAD!
             if $some_error_condition;

       It should use "postpone":

          AnyEvent::postpone { $cb->(undef) }, return # signal error to callback, later
             if $some_error_condition;

   AnyEvent::log $level, $msg[, @args]
       Log the given $msg at the given $level.

       If AnyEvent::Log is not loaded then this function makes a simple
       test to see whether the message will be logged. If the test succeeds
       it will load AnyEvent::Log and call "AnyEvent::Log::log" -
       consequently, look at the AnyEvent::Log documentation for details.

       If the test fails it will simply return. Right now this happens when
       a numerical loglevel is used and it is larger than the level
       specified via $ENV{PERL_ANYEVENT_VERBOSE}.

       If you want to sprinkle loads of logging calls around your code,
       consider creating a logger callback with the "AnyEvent::Log::logger"
       function, which can reduce typing, codesize and can reduce the
       logging overhead enourmously.

   AnyEvent::fh_block $filehandle
   AnyEvent::fh_unblock $filehandle
       Sets blocking or non-blocking behaviour for the given filehandle.

WHAT TO DO IN A MODULE
   As a module author, you should "use AnyEvent" and call AnyEvent methods
   freely, but you should not load a specific event module or rely on it.

   Be careful when you create watchers in the module body - AnyEvent will
   decide which event module to use as soon as the first method is called,
   so by calling AnyEvent in your module body you force the user of your
   module to load the event module first.

   Never call "->recv" on a condition variable unless you *know* that the
   "->send" method has been called on it already. This is because it will
   stall the whole program, and the whole point of using events is to stay
   interactive.

   It is fine, however, to call "->recv" when the user of your module
   requests it (i.e. if you create a http request object ad have a method
   called "results" that returns the results, it may call "->recv" freely,
   as the user of your module knows what she is doing. Always).

WHAT TO DO IN THE MAIN PROGRAM
   There will always be a single main program - the only place that should
   dictate which event model to use.

   If the program is not event-based, it need not do anything special, even
   when it depends on a module that uses an AnyEvent. If the program itself
   uses AnyEvent, but does not care which event loop is used, all it needs
   to do is "use AnyEvent". In either case, AnyEvent will choose the best
   available loop implementation.

   If the main program relies on a specific event model - for example, in
   Gtk2 programs you have to rely on the Glib module - you should load the
   event module before loading AnyEvent or any module that uses it:
   generally speaking, you should load it as early as possible. The reason
   is that modules might create watchers when they are loaded, and AnyEvent
   will decide on the event model to use as soon as it creates watchers,
   and it might choose the wrong one unless you load the correct one
   yourself.

   You can chose to use a pure-perl implementation by loading the
   "AnyEvent::Loop" module, which gives you similar behaviour everywhere,
   but letting AnyEvent chose the model is generally better.

 MAINLOOP EMULATION
   Sometimes (often for short test scripts, or even standalone programs who
   only want to use AnyEvent), you do not want to run a specific event
   loop.

   In that case, you can use a condition variable like this:

      AnyEvent->condvar->recv;

   This has the effect of entering the event loop and looping forever.

   Note that usually your program has some exit condition, in which case it
   is better to use the "traditional" approach of storing a condition
   variable somewhere, waiting for it, and sending it when the program
   should exit cleanly.

OTHER MODULES
   The following is a non-exhaustive list of additional modules that use
   AnyEvent as a client and can therefore be mixed easily with other
   AnyEvent modules and other event loops in the same program. Some of the
   modules come as part of AnyEvent, the others are available via CPAN (see
   <http://search.cpan.org/search?m=module&q=anyevent%3A%3A*> for a longer
   non-exhaustive list), and the list is heavily biased towards modules of
   the AnyEvent author himself :)

   AnyEvent::Util (part of the AnyEvent distribution)
       Contains various utility functions that replace often-used blocking
       functions such as "inet_aton" with event/callback-based versions.

   AnyEvent::Socket (part of the AnyEvent distribution)
       Provides various utility functions for (internet protocol) sockets,
       addresses and name resolution. Also functions to create non-blocking
       tcp connections or tcp servers, with IPv6 and SRV record support and
       more.

   AnyEvent::Handle (part of the AnyEvent distribution)
       Provide read and write buffers, manages watchers for reads and
       writes, supports raw and formatted I/O, I/O queued and fully
       transparent and non-blocking SSL/TLS (via AnyEvent::TLS).

   AnyEvent::DNS (part of the AnyEvent distribution)
       Provides rich asynchronous DNS resolver capabilities.

   AnyEvent::HTTP, AnyEvent::IRC, AnyEvent::XMPP, AnyEvent::GPSD,
   AnyEvent::IGS, AnyEvent::FCP
       Implement event-based interfaces to the protocols of the same name
       (for the curious, IGS is the International Go Server and FCP is the
       Freenet Client Protocol).

   AnyEvent::AIO (part of the AnyEvent distribution)
       Truly asynchronous (as opposed to non-blocking) I/O, should be in
       the toolbox of every event programmer. AnyEvent::AIO transparently
       fuses IO::AIO and AnyEvent together, giving AnyEvent access to
       event-based file I/O, and much more.

   AnyEvent::Fork, AnyEvent::Fork::RPC, AnyEvent::Fork::Pool,
   AnyEvent::Fork::Remote
       These let you safely fork new subprocesses, either locally or
       remotely (e.g.v ia ssh), using some RPC protocol or not, without the
       limitations normally imposed by fork (AnyEvent works fine for
       example). Dynamically-resized worker pools are obviously included as
       well.

       And they are quite tiny and fast as well - "abusing" AnyEvent::Fork
       just to exec external programs can easily beat using "fork" and
       "exec" (or even "system") in most programs.

   AnyEvent::Filesys::Notify
       AnyEvent is good for non-blocking stuff, but it can't detect file or
       path changes (e.g. "watch this directory for new files", "watch this
       file for changes"). The AnyEvent::Filesys::Notify module promises to
       do just that in a portbale fashion, supporting inotify on GNU/Linux
       and some weird, without doubt broken, stuff on OS X to monitor
       files. It can fall back to blocking scans at regular intervals
       transparently on other platforms, so it's about as portable as it
       gets.

       (I haven't used it myself, but it seems the biggest problem with it
       is it quite bad performance).

   AnyEvent::DBI
       Executes DBI requests asynchronously in a proxy process for you,
       notifying you in an event-based way when the operation is finished.

   AnyEvent::FastPing
       The fastest ping in the west.

   Coro
       Has special support for AnyEvent via Coro::AnyEvent, which allows
       you to simply invert the flow control - don't call us, we will call
       you:

          async {
             Coro::AnyEvent::sleep 5; # creates a 5s timer and waits for it
             print "5 seconds later!\n";

             Coro::AnyEvent::readable *STDIN; # uses an I/O watcher
             my $line = <STDIN>; # works for ttys

             AnyEvent::HTTP::http_get "url", Coro::rouse_cb;
             my ($body, $hdr) = Coro::rouse_wait;
          };

SIMPLIFIED AE API
   Starting with version 5.0, AnyEvent officially supports a second, much
   simpler, API that is designed to reduce the calling, typing and memory
   overhead by using function call syntax and a fixed number of parameters.

   See the AE manpage for details.

ERROR AND EXCEPTION HANDLING
   In general, AnyEvent does not do any error handling - it relies on the
   caller to do that if required. The AnyEvent::Strict module (see also the
   "PERL_ANYEVENT_STRICT" environment variable, below) provides strict
   checking of all AnyEvent methods, however, which is highly useful during
   development.

   As for exception handling (i.e. runtime errors and exceptions thrown
   while executing a callback), this is not only highly event-loop
   specific, but also not in any way wrapped by this module, as this is the
   job of the main program.

   The pure perl event loop simply re-throws the exception (usually within
   "condvar->recv"), the Event and EV modules call "$Event/EV::DIED->()",
   Glib uses "install_exception_handler" and so on.

ENVIRONMENT VARIABLES
   AnyEvent supports a number of environment variables that tune the
   runtime behaviour. They are usually evaluated when AnyEvent is loaded,
   initialised, or a submodule that uses them is loaded. Many of them also
   cause AnyEvent to load additional modules - for example,
   "PERL_ANYEVENT_DEBUG_WRAP" causes the AnyEvent::Debug module to be
   loaded.

   All the environment variables documented here start with
   "PERL_ANYEVENT_", which is what AnyEvent considers its own namespace.
   Other modules are encouraged (but by no means required) to use
   "PERL_ANYEVENT_SUBMODULE" if they have registered the
   AnyEvent::Submodule namespace on CPAN, for any submodule. For example,
   AnyEvent::HTTP could be expected to use "PERL_ANYEVENT_HTTP_PROXY" (it
   should not access env variables starting with "AE_", see below).

   All variables can also be set via the "AE_" prefix, that is, instead of
   setting "PERL_ANYEVENT_VERBOSE" you can also set "AE_VERBOSE". In case
   there is a clash btween anyevent and another program that uses
   "AE_something" you can set the corresponding "PERL_ANYEVENT_something"
   variable to the empty string, as those variables take precedence.

   When AnyEvent is first loaded, it copies all "AE_xxx" env variables to
   their "PERL_ANYEVENT_xxx" counterpart unless that variable already
   exists. If taint mode is on, then AnyEvent will remove *all* environment
   variables starting with "PERL_ANYEVENT_" from %ENV (or replace them with
   "undef" or the empty string, if the corresaponding "AE_" variable is
   set).

   The exact algorithm is currently:

      1. if taint mode enabled, delete all PERL_ANYEVENT_xyz variables from %ENV
      2. copy over AE_xyz to PERL_ANYEVENT_xyz unless the latter alraedy exists
      3. if taint mode enabled, set all PERL_ANYEVENT_xyz variables to undef.

   This ensures that child processes will not see the "AE_" variables.

   The following environment variables are currently known to AnyEvent:

   "PERL_ANYEVENT_VERBOSE"
       By default, AnyEvent will log messages with loglevel 4 ("error") or
       higher (see AnyEvent::Log). You can set this environment variable to
       a numerical loglevel to make AnyEvent more (or less) talkative.

       If you want to do more than just set the global logging level you
       should have a look at "PERL_ANYEVENT_LOG", which allows much more
       complex specifications.

       When set to 0 ("off"), then no messages whatsoever will be logged
       with everything else at defaults.

       When set to 5 or higher ("warn"), AnyEvent warns about unexpected
       conditions, such as not being able to load the event model specified
       by "PERL_ANYEVENT_MODEL", or a guard callback throwing an exception
       - this is the minimum recommended level for use during development.

       When set to 7 or higher (info), AnyEvent reports which event model
       it chooses.

       When set to 8 or higher (debug), then AnyEvent will report extra
       information on which optional modules it loads and how it implements
       certain features.

   "PERL_ANYEVENT_LOG"
       Accepts rather complex logging specifications. For example, you
       could log all "debug" messages of some module to stderr, warnings
       and above to stderr, and errors and above to syslog, with:

          PERL_ANYEVENT_LOG=Some::Module=debug,+log:filter=warn,+%syslog:%syslog=error,syslog

       For the rather extensive details, see AnyEvent::Log.

       This variable is evaluated when AnyEvent (or AnyEvent::Log) is
       loaded, so will take effect even before AnyEvent has initialised
       itself.

       Note that specifying this environment variable causes the
       AnyEvent::Log module to be loaded, while "PERL_ANYEVENT_VERBOSE"
       does not, so only using the latter saves a few hundred kB of memory
       unless a module explicitly needs the extra features of
       AnyEvent::Log.

   "PERL_ANYEVENT_STRICT"
       AnyEvent does not do much argument checking by default, as thorough
       argument checking is very costly. Setting this variable to a true
       value will cause AnyEvent to load "AnyEvent::Strict" and then to
       thoroughly check the arguments passed to most method calls. If it
       finds any problems, it will croak.

       In other words, enables "strict" mode.

       Unlike "use strict" (or its modern cousin, "use common::sense", it
       is definitely recommended to keep it off in production. Keeping
       "PERL_ANYEVENT_STRICT=1" in your environment while developing
       programs can be very useful, however.

   "PERL_ANYEVENT_DEBUG_SHELL"
       If this env variable is nonempty, then its contents will be
       interpreted by "AnyEvent::Socket::parse_hostport" and
       "AnyEvent::Debug::shell" (after replacing every occurance of $$ by
       the process pid). The shell object is saved in
       $AnyEvent::Debug::SHELL.

       This happens when the first watcher is created.

       For example, to bind a debug shell on a unix domain socket in
       /tmp/debug<pid>.sock, you could use this:

          PERL_ANYEVENT_DEBUG_SHELL=/tmp/debug\$\$.sock perlprog
          # connect with e.g.: socat readline /tmp/debug123.sock

       Or to bind to tcp port 4545 on localhost:

          PERL_ANYEVENT_DEBUG_SHELL=127.0.0.1:4545 perlprog
          # connect with e.g.: telnet localhost 4545

       Note that creating sockets in /tmp or on localhost is very unsafe on
       multiuser systems.

   "PERL_ANYEVENT_DEBUG_WRAP"
       Can be set to 0, 1 or 2 and enables wrapping of all watchers for
       debugging purposes. See "AnyEvent::Debug::wrap" for details.

   "PERL_ANYEVENT_MODEL"
       This can be used to specify the event model to be used by AnyEvent,
       before auto detection and -probing kicks in.

       It normally is a string consisting entirely of ASCII letters (e.g.
       "EV" or "IOAsync"). The string "AnyEvent::Impl::" gets prepended and
       the resulting module name is loaded and - if the load was successful
       - used as event model backend. If it fails to load then AnyEvent
       will proceed with auto detection and -probing.

       If the string ends with "::" instead (e.g. "AnyEvent::Impl::EV::")
       then nothing gets prepended and the module name is used as-is (hint:
       "::" at the end of a string designates a module name and quotes it
       appropriately).

       For example, to force the pure perl model (AnyEvent::Loop::Perl) you
       could start your program like this:

          PERL_ANYEVENT_MODEL=Perl perl ...

   "PERL_ANYEVENT_IO_MODEL"
       The current file I/O model - see AnyEvent::IO for more info.

       At the moment, only "Perl" (small, pure-perl, synchronous) and
       "IOAIO" (truly asynchronous) are supported. The default is "IOAIO"
       if AnyEvent::AIO can be loaded, otherwise it is "Perl".

   "PERL_ANYEVENT_PROTOCOLS"
       Used by both AnyEvent::DNS and AnyEvent::Socket to determine
       preferences for IPv4 or IPv6. The default is unspecified (and might
       change, or be the result of auto probing).

       Must be set to a comma-separated list of protocols or address
       families, current supported: "ipv4" and "ipv6". Only protocols
       mentioned will be used, and preference will be given to protocols
       mentioned earlier in the list.

       This variable can effectively be used for denial-of-service attacks
       against local programs (e.g. when setuid), although the impact is
       likely small, as the program has to handle connection and other
       failures anyways.

       Examples: "PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6" - prefer IPv4 over
       IPv6, but support both and try to use both.
       "PERL_ANYEVENT_PROTOCOLS=ipv4" - only support IPv4, never try to
       resolve or contact IPv6 addresses.
       "PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4" support either IPv4 or IPv6, but
       prefer IPv6 over IPv4.

   "PERL_ANYEVENT_HOSTS"
       This variable, if specified, overrides the /etc/hosts file used by
       AnyEvent::Socket"::resolve_sockaddr", i.e. hosts aliases will be
       read from that file instead.

   "PERL_ANYEVENT_EDNS0"
       Used by AnyEvent::DNS to decide whether to use the EDNS0 extension
       for DNS. This extension is generally useful to reduce DNS traffic,
       especially when DNSSEC is involved, but some (broken) firewalls drop
       such DNS packets, which is why it is off by default.

       Setting this variable to 1 will cause AnyEvent::DNS to announce
       EDNS0 in its DNS requests.

   "PERL_ANYEVENT_MAX_FORKS"
       The maximum number of child processes that
       "AnyEvent::Util::fork_call" will create in parallel.

   "PERL_ANYEVENT_MAX_OUTSTANDING_DNS"
       The default value for the "max_outstanding" parameter for the
       default DNS resolver - this is the maximum number of parallel DNS
       requests that are sent to the DNS server.

   "PERL_ANYEVENT_MAX_SIGNAL_LATENCY"
       Perl has inherently racy signal handling (you can basically choose
       between losing signals and memory corruption) - pure perl event
       loops (including "AnyEvent::Loop", when "Async::Interrupt" isn't
       available) therefore have to poll regularly to avoid losing signals.

       Some event loops are racy, but don't poll regularly, and some event
       loops are written in C but are still racy. For those event loops,
       AnyEvent installs a timer that regularly wakes up the event loop.

       By default, the interval for this timer is 10 seconds, but you can
       override this delay with this environment variable (or by setting
       the $AnyEvent::MAX_SIGNAL_LATENCY variable before creating signal
       watchers).

       Lower values increase CPU (and energy) usage, higher values can
       introduce long delays when reaping children or waiting for signals.

       The AnyEvent::Async module, if available, will be used to avoid this
       polling (with most event loops).

   "PERL_ANYEVENT_RESOLV_CONF"
       The absolute path to a resolv.conf-style file to use instead of
       /etc/resolv.conf (or the OS-specific configuration) in the default
       resolver, or the empty string to select the default configuration.

   "PERL_ANYEVENT_CA_FILE", "PERL_ANYEVENT_CA_PATH".
       When neither "ca_file" nor "ca_path" was specified during
       AnyEvent::TLS context creation, and either of these environment
       variables are nonempty, they will be used to specify CA certificate
       locations instead of a system-dependent default.

   "PERL_ANYEVENT_AVOID_GUARD" and "PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT"
       When these are set to 1, then the respective modules are not loaded.
       Mostly good for testing AnyEvent itself.

SUPPLYING YOUR OWN EVENT MODEL INTERFACE
   This is an advanced topic that you do not normally need to use AnyEvent
   in a module. This section is only of use to event loop authors who want
   to provide AnyEvent compatibility.

   If you need to support another event library which isn't directly
   supported by AnyEvent, you can supply your own interface to it by
   pushing, before the first watcher gets created, the package name of the
   event module and the package name of the interface to use onto
   @AnyEvent::REGISTRY. You can do that before and even without loading
   AnyEvent, so it is reasonably cheap.

   Example:

      push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::];

   This tells AnyEvent to (literally) use the "urxvt::anyevent::"
   package/class when it finds the "urxvt" package/module is already
   loaded.

   When AnyEvent is loaded and asked to find a suitable event model, it
   will first check for the presence of urxvt by trying to "use" the
   "urxvt::anyevent" module.

   The class should provide implementations for all watcher types. See
   AnyEvent::Impl::EV (source code), AnyEvent::Impl::Glib (Source code) and
   so on for actual examples. Use "perldoc -m AnyEvent::Impl::Glib" to see
   the sources.

   If you don't provide "signal" and "child" watchers than AnyEvent will
   provide suitable (hopefully) replacements.

   The above example isn't fictitious, the *rxvt-unicode* (a.k.a. urxvt)
   terminal emulator uses the above line as-is. An interface isn't included
   in AnyEvent because it doesn't make sense outside the embedded
   interpreter inside *rxvt-unicode*, and it is updated and maintained as
   part of the *rxvt-unicode* distribution.

   *rxvt-unicode* also cheats a bit by not providing blocking access to
   condition variables: code blocking while waiting for a condition will
   "die". This still works with most modules/usages, and blocking calls
   must not be done in an interactive application, so it makes sense.

EXAMPLE PROGRAM
   The following program uses an I/O watcher to read data from STDIN, a
   timer to display a message once per second, and a condition variable to
   quit the program when the user enters quit:

      use AnyEvent;

      my $cv = AnyEvent->condvar;

      my $io_watcher = AnyEvent->io (
         fh   => \*STDIN,
         poll => 'r',
         cb   => sub {
            warn "io event <$_[0]>\n";   # will always output <r>
            chomp (my $input = <STDIN>); # read a line
            warn "read: $input\n";       # output what has been read
            $cv->send if $input =~ /^q/i; # quit program if /^q/i
         },
      );

      my $time_watcher = AnyEvent->timer (after => 1, interval => 1, cb => sub {
         warn "timeout\n"; # print 'timeout' at most every second
      });

      $cv->recv; # wait until user enters /^q/i

REAL-WORLD EXAMPLE
   Consider the Net::FCP module. It features (among others) the following
   API calls, which are to freenet what HTTP GET requests are to http:

      my $data = $fcp->client_get ($url); # blocks

      my $transaction = $fcp->txn_client_get ($url); # does not block
      $transaction->cb ( sub { ... } ); # set optional result callback
      my $data = $transaction->result; # possibly blocks

   The "client_get" method works like "LWP::Simple::get": it requests the
   given URL and waits till the data has arrived. It is defined to be:

      sub client_get { $_[0]->txn_client_get ($_[1])->result }

   And in fact is automatically generated. This is the blocking API of
   Net::FCP, and it works as simple as in any other, similar, module.

   More complicated is "txn_client_get": It only creates a transaction
   (completion, result, ...) object and initiates the transaction.

      my $txn = bless { }, Net::FCP::Txn::;

   It also creates a condition variable that is used to signal the
   completion of the request:

      $txn->{finished} = AnyAvent->condvar;

   It then creates a socket in non-blocking mode.

      socket $txn->{fh}, ...;
      fcntl $txn->{fh}, F_SETFL, O_NONBLOCK;
      connect $txn->{fh}, ...
         and !$!{EWOULDBLOCK}
         and !$!{EINPROGRESS}
         and Carp::croak "unable to connect: $!\n";

   Then it creates a write-watcher which gets called whenever an error
   occurs or the connection succeeds:

      $txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w });

   And returns this transaction object. The "fh_ready_w" callback gets
   called as soon as the event loop detects that the socket is ready for
   writing.

   The "fh_ready_w" method makes the socket blocking again, writes the
   request data and replaces the watcher by a read watcher (waiting for
   reply data). The actual code is more complicated, but that doesn't
   matter for this example:

      fcntl $txn->{fh}, F_SETFL, 0;
      syswrite $txn->{fh}, $txn->{request}
         or die "connection or write error";
      $txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r });

   Again, "fh_ready_r" waits till all data has arrived, and then stores the
   result and signals any possible waiters that the request has finished:

      sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf};

      if (end-of-file or data complete) {
        $txn->{result} = $txn->{buf};
        $txn->{finished}->send;
        $txb->{cb}->($txn) of $txn->{cb}; # also call callback
      }

   The "result" method, finally, just waits for the finished signal (if the
   request was already finished, it doesn't wait, of course, and returns
   the data:

      $txn->{finished}->recv;
      return $txn->{result};

   The actual code goes further and collects all errors ("die"s,
   exceptions) that occurred during request processing. The "result" method
   detects whether an exception as thrown (it is stored inside the $txn
   object) and just throws the exception, which means connection errors and
   other problems get reported to the code that tries to use the result,
   not in a random callback.

   All of this enables the following usage styles:

   1. Blocking:

      my $data = $fcp->client_get ($url);

   2. Blocking, but running in parallel:

      my @datas = map $_->result,
                     map $fcp->txn_client_get ($_),
                        @urls;

   Both blocking examples work without the module user having to know
   anything about events.

   3a. Event-based in a main program, using any supported event module:

      use EV;

      $fcp->txn_client_get ($url)->cb (sub {
         my $txn = shift;
         my $data = $txn->result;
         ...
      });

      EV::run;

   3b. The module user could use AnyEvent, too:

      use AnyEvent;

      my $quit = AnyEvent->condvar;

      $fcp->txn_client_get ($url)->cb (sub {
         ...
         $quit->send;
      });

      $quit->recv;

BENCHMARKS
   To give you an idea of the performance and overheads that AnyEvent adds
   over the event loops themselves and to give you an impression of the
   speed of various event loops I prepared some benchmarks.

 BENCHMARKING ANYEVENT OVERHEAD
   Here is a benchmark of various supported event models used natively and
   through AnyEvent. The benchmark creates a lot of timers (with a zero
   timeout) and I/O watchers (watching STDOUT, a pty, to become writable,
   which it is), lets them fire exactly once and destroys them again.

   Source code for this benchmark is found as eg/bench in the AnyEvent
   distribution. It uses the AE interface, which makes a real difference
   for the EV and Perl backends only.

  Explanation of the columns
   *watcher* is the number of event watchers created/destroyed. Since
   different event models feature vastly different performances, each event
   loop was given a number of watchers so that overall runtime is
   acceptable and similar between tested event loop (and keep them from
   crashing): Glib would probably take thousands of years if asked to
   process the same number of watchers as EV in this benchmark.

   *bytes* is the number of bytes (as measured by the resident set size,
   RSS) consumed by each watcher. This method of measuring captures both C
   and Perl-based overheads.

   *create* is the time, in microseconds (millionths of seconds), that it
   takes to create a single watcher. The callback is a closure shared
   between all watchers, to avoid adding memory overhead. That means
   closure creation and memory usage is not included in the figures.

   *invoke* is the time, in microseconds, used to invoke a simple callback.
   The callback simply counts down a Perl variable and after it was invoked
   "watcher" times, it would "->send" a condvar once to signal the end of
   this phase.

   *destroy* is the time, in microseconds, that it takes to destroy a
   single watcher.

  Results
             name watchers bytes create invoke destroy comment
            EV/EV   100000   223   0.47   0.43    0.27 EV native interface
           EV/Any   100000   223   0.48   0.42    0.26 EV + AnyEvent watchers
     Coro::EV/Any   100000   223   0.47   0.42    0.26 coroutines + Coro::Signal
         Perl/Any   100000   431   2.70   0.74    0.92 pure perl implementation
      Event/Event    16000   516  31.16  31.84    0.82 Event native interface
        Event/Any    16000  1203  42.61  34.79    1.80 Event + AnyEvent watchers
      IOAsync/Any    16000  1911  41.92  27.45   16.81 via IO::Async::Loop::IO_Poll
      IOAsync/Any    16000  1726  40.69  26.37   15.25 via IO::Async::Loop::Epoll
         Glib/Any    16000  1118  89.00  12.57   51.17 quadratic behaviour
           Tk/Any     2000  1346  20.96  10.75    8.00 SEGV with >> 2000 watchers
          POE/Any     2000  6951 108.97 795.32   14.24 via POE::Loop::Event
          POE/Any     2000  6648  94.79 774.40  575.51 via POE::Loop::Select

  Discussion
   The benchmark does *not* measure scalability of the event loop very
   well. For example, a select-based event loop (such as the pure perl one)
   can never compete with an event loop that uses epoll when the number of
   file descriptors grows high. In this benchmark, all events become ready
   at the same time, so select/poll-based implementations get an unnatural
   speed boost.

   Also, note that the number of watchers usually has a nonlinear effect on
   overall speed, that is, creating twice as many watchers doesn't take
   twice the time - usually it takes longer. This puts event loops tested
   with a higher number of watchers at a disadvantage.

   To put the range of results into perspective, consider that on the
   benchmark machine, handling an event takes roughly 1600 CPU cycles with
   EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000
   CPU cycles with POE.

   "EV" is the sole leader regarding speed and memory use, which are both
   maximal/minimal, respectively. When using the AE API there is zero
   overhead (when going through the AnyEvent API create is about 5-6 times
   slower, with other times being equal, so still uses far less memory than
   any other event loop and is still faster than Event natively).

   The pure perl implementation is hit in a few sweet spots (both the
   constant timeout and the use of a single fd hit optimisations in the
   perl interpreter and the backend itself). Nevertheless this shows that
   it adds very little overhead in itself. Like any select-based backend
   its performance becomes really bad with lots of file descriptors (and
   few of them active), of course, but this was not subject of this
   benchmark.

   The "Event" module has a relatively high setup and callback invocation
   cost, but overall scores in on the third place.

   "IO::Async" performs admirably well, about on par with "Event", even
   when using its pure perl backend.

   "Glib"'s memory usage is quite a bit higher, but it features a faster
   callback invocation and overall ends up in the same class as "Event".
   However, Glib scales extremely badly, doubling the number of watchers
   increases the processing time by more than a factor of four, making it
   completely unusable when using larger numbers of watchers (note that
   only a single file descriptor was used in the benchmark, so
   inefficiencies of "poll" do not account for this).

   The "Tk" adaptor works relatively well. The fact that it crashes with
   more than 2000 watchers is a big setback, however, as correctness takes
   precedence over speed. Nevertheless, its performance is surprising, as
   the file descriptor is dup()ed for each watcher. This shows that the
   dup() employed by some adaptors is not a big performance issue (it does
   incur a hidden memory cost inside the kernel which is not reflected in
   the figures above).

   "POE", regardless of underlying event loop (whether using its pure perl
   select-based backend or the Event module, the POE-EV backend couldn't be
   tested because it wasn't working) shows abysmal performance and memory
   usage with AnyEvent: Watchers use almost 30 times as much memory as EV
   watchers, and 10 times as much memory as Event (the high memory
   requirements are caused by requiring a session for each watcher).
   Watcher invocation speed is almost 900 times slower than with AnyEvent's
   pure perl implementation.

   The design of the POE adaptor class in AnyEvent can not really account
   for the performance issues, though, as session creation overhead is
   small compared to execution of the state machine, which is coded pretty
   optimally within AnyEvent::Impl::POE (and while everybody agrees that
   using multiple sessions is not a good approach, especially regarding
   memory usage, even the author of POE could not come up with a faster
   design).

  Summary
   *   Using EV through AnyEvent is faster than any other event loop (even
       when used without AnyEvent), but most event loops have acceptable
       performance with or without AnyEvent.

   *   The overhead AnyEvent adds is usually much smaller than the overhead
       of the actual event loop, only with extremely fast event loops such
       as EV does AnyEvent add significant overhead.

   *   You should avoid POE like the plague if you want performance or
       reasonable memory usage.

 BENCHMARKING THE LARGE SERVER CASE
   This benchmark actually benchmarks the event loop itself. It works by
   creating a number of "servers": each server consists of a socket pair, a
   timeout watcher that gets reset on activity (but never fires), and an
   I/O watcher waiting for input on one side of the socket. Each time the
   socket watcher reads a byte it will write that byte to a random other
   "server".

   The effect is that there will be a lot of I/O watchers, only part of
   which are active at any one point (so there is a constant number of
   active fds for each loop iteration, but which fds these are is random).
   The timeout is reset each time something is read because that reflects
   how most timeouts work (and puts extra pressure on the event loops).

   In this benchmark, we use 10000 socket pairs (20000 sockets), of which
   100 (1%) are active. This mirrors the activity of large servers with
   many connections, most of which are idle at any one point in time.

   Source code for this benchmark is found as eg/bench2 in the AnyEvent
   distribution. It uses the AE interface, which makes a real difference
   for the EV and Perl backends only.

  Explanation of the columns
   *sockets* is the number of sockets, and twice the number of "servers"
   (as each server has a read and write socket end).

   *create* is the time it takes to create a socket pair (which is
   nontrivial) and two watchers: an I/O watcher and a timeout watcher.

   *request*, the most important value, is the time it takes to handle a
   single "request", that is, reading the token from the pipe and
   forwarding it to another server. This includes deleting the old timeout
   and creating a new one that moves the timeout into the future.

  Results
        name sockets create  request
          EV   20000  62.66     7.99
        Perl   20000  68.32    32.64
     IOAsync   20000 174.06   101.15 epoll
     IOAsync   20000 174.67   610.84 poll
       Event   20000 202.69   242.91
        Glib   20000 557.01  1689.52
         POE   20000 341.54 12086.32 uses POE::Loop::Event

  Discussion
   This benchmark *does* measure scalability and overall performance of the
   particular event loop.

   EV is again fastest. Since it is using epoll on my system, the setup
   time is relatively high, though.

   Perl surprisingly comes second. It is much faster than the C-based event
   loops Event and Glib.

   IO::Async performs very well when using its epoll backend, and still
   quite good compared to Glib when using its pure perl backend.

   Event suffers from high setup time as well (look at its code and you
   will understand why). Callback invocation also has a high overhead
   compared to the "$_->() for .."-style loop that the Perl event loop
   uses. Event uses select or poll in basically all documented
   configurations.

   Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It
   clearly fails to perform with many filehandles or in busy servers.

   POE is still completely out of the picture, taking over 1000 times as
   long as EV, and over 100 times as long as the Perl implementation, even
   though it uses a C-based event loop in this case.

  Summary
   *   The pure perl implementation performs extremely well.

   *   Avoid Glib or POE in large projects where performance matters.

 BENCHMARKING SMALL SERVERS
   While event loops should scale (and select-based ones do not...) even to
   large servers, most programs we (or I :) actually write have only a few
   I/O watchers.

   In this benchmark, I use the same benchmark program as in the large
   server case, but it uses only eight "servers", of which three are active
   at any one time. This should reflect performance for a small server
   relatively well.

   The columns are identical to the previous table.

  Results
       name sockets create request
         EV      16  20.00    6.54
       Perl      16  25.75   12.62
      Event      16  81.27   35.86
       Glib      16  32.63   15.48
        POE      16 261.87  276.28 uses POE::Loop::Event

  Discussion
   The benchmark tries to test the performance of a typical small server.
   While knowing how various event loops perform is interesting, keep in
   mind that their overhead in this case is usually not as important, due
   to the small absolute number of watchers (that is, you need efficiency
   and speed most when you have lots of watchers, not when you only have a
   few of them).

   EV is again fastest.

   Perl again comes second. It is noticeably faster than the C-based event
   loops Event and Glib, although the difference is too small to really
   matter.

   POE also performs much better in this case, but is is still far behind
   the others.

  Summary
   *   C-based event loops perform very well with small number of watchers,
       as the management overhead dominates.

 THE IO::Lambda BENCHMARK
   Recently I was told about the benchmark in the IO::Lambda manpage, which
   could be misinterpreted to make AnyEvent look bad. In fact, the
   benchmark simply compares IO::Lambda with POE, and IO::Lambda looks
   better (which shouldn't come as a surprise to anybody). As such, the
   benchmark is fine, and mostly shows that the AnyEvent backend from
   IO::Lambda isn't very optimal. But how would AnyEvent compare when used
   without the extra baggage? To explore this, I wrote the equivalent
   benchmark for AnyEvent.

   The benchmark itself creates an echo-server, and then, for 500 times,
   connects to the echo server, sends a line, waits for the reply, and then
   creates the next connection. This is a rather bad benchmark, as it
   doesn't test the efficiency of the framework or much non-blocking I/O,
   but it is a benchmark nevertheless.

      name                    runtime
      Lambda/select           0.330 sec
         + optimized          0.122 sec
      Lambda/AnyEvent         0.327 sec
         + optimized          0.138 sec
      Raw sockets/select      0.077 sec
      POE/select, components  0.662 sec
      POE/select, raw sockets 0.226 sec
      POE/select, optimized   0.404 sec

      AnyEvent/select/nb      0.085 sec
      AnyEvent/EV/nb          0.068 sec
         +state machine       0.134 sec

   The benchmark is also a bit unfair (my fault): the IO::Lambda/POE
   benchmarks actually make blocking connects and use 100% blocking I/O,
   defeating the purpose of an event-based solution. All of the newly
   written AnyEvent benchmarks use 100% non-blocking connects (using
   AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS
   resolver), so AnyEvent is at a disadvantage here, as non-blocking
   connects generally require a lot more bookkeeping and event handling
   than blocking connects (which involve a single syscall only).

   The last AnyEvent benchmark additionally uses AnyEvent::Handle, which
   offers similar expressive power as POE and IO::Lambda, using
   conventional Perl syntax. This means that both the echo server and the
   client are 100% non-blocking, further placing it at a disadvantage.

   As you can see, the AnyEvent + EV combination even beats the
   hand-optimised "raw sockets benchmark", while AnyEvent + its pure perl
   backend easily beats IO::Lambda and POE.

   And even the 100% non-blocking version written using the high-level (and
   slow :) AnyEvent::Handle abstraction beats both POE and IO::Lambda
   higher level ("unoptimised") abstractions by a large margin, even though
   it does all of DNS, tcp-connect and socket I/O in a non-blocking way.

   The two AnyEvent benchmarks programs can be found as eg/ae0.pl and
   eg/ae2.pl in the AnyEvent distribution, the remaining benchmarks are
   part of the IO::Lambda distribution and were used without any changes.

SIGNALS
   AnyEvent currently installs handlers for these signals:

   SIGCHLD
       A handler for "SIGCHLD" is installed by AnyEvent's child watcher
       emulation for event loops that do not support them natively. Also,
       some event loops install a similar handler.

       Additionally, when AnyEvent is loaded and SIGCHLD is set to IGNORE,
       then AnyEvent will reset it to default, to avoid losing child exit
       statuses.

   SIGPIPE
       A no-op handler is installed for "SIGPIPE" when $SIG{PIPE} is
       "undef" when AnyEvent gets loaded.

       The rationale for this is that AnyEvent users usually do not really
       depend on SIGPIPE delivery (which is purely an optimisation for
       shell use, or badly-written programs), but "SIGPIPE" can cause
       spurious and rare program exits as a lot of people do not expect
       "SIGPIPE" when writing to some random socket.

       The rationale for installing a no-op handler as opposed to ignoring
       it is that this way, the handler will be restored to defaults on
       exec.

       Feel free to install your own handler, or reset it to defaults.

RECOMMENDED/OPTIONAL MODULES
   One of AnyEvent's main goals is to be 100% Pure-Perl(tm): only perl (and
   its built-in modules) are required to use it.

   That does not mean that AnyEvent won't take advantage of some additional
   modules if they are installed.

   This section explains which additional modules will be used, and how
   they affect AnyEvent's operation.

   Async::Interrupt
       This slightly arcane module is used to implement fast signal
       handling: To my knowledge, there is no way to do completely
       race-free and quick signal handling in pure perl. To ensure that
       signals still get delivered, AnyEvent will start an interval timer
       to wake up perl (and catch the signals) with some delay (default is
       10 seconds, look for $AnyEvent::MAX_SIGNAL_LATENCY).

       If this module is available, then it will be used to implement
       signal catching, which means that signals will not be delayed, and
       the event loop will not be interrupted regularly, which is more
       efficient (and good for battery life on laptops).

       This affects not just the pure-perl event loop, but also other event
       loops that have no signal handling on their own (e.g. Glib, Tk, Qt).

       Some event loops (POE, Event, Event::Lib) offer signal watchers
       natively, and either employ their own workarounds (POE) or use
       AnyEvent's workaround (using $AnyEvent::MAX_SIGNAL_LATENCY).
       Installing Async::Interrupt does nothing for those backends.

   EV  This module isn't really "optional", as it is simply one of the
       backend event loops that AnyEvent can use. However, it is simply the
       best event loop available in terms of features, speed and stability:
       It supports the AnyEvent API optimally, implements all the watcher
       types in XS, does automatic timer adjustments even when no monotonic
       clock is available, can take avdantage of advanced kernel interfaces
       such as "epoll" and "kqueue", and is the fastest backend *by far*.
       You can even embed Glib/Gtk2 in it (or vice versa, see EV::Glib and
       Glib::EV).

       If you only use backends that rely on another event loop (e.g.
       "Tk"), then this module will do nothing for you.

   Guard
       The guard module, when used, will be used to implement
       "AnyEvent::Util::guard". This speeds up guards considerably (and
       uses a lot less memory), but otherwise doesn't affect guard
       operation much. It is purely used for performance.

   JSON and JSON::XS
       One of these modules is required when you want to read or write JSON
       data via AnyEvent::Handle. JSON is also written in pure-perl, but
       can take advantage of the ultra-high-speed JSON::XS module when it
       is installed.

   Net::SSLeay
       Implementing TLS/SSL in Perl is certainly interesting, but not very
       worthwhile: If this module is installed, then AnyEvent::Handle (with
       the help of AnyEvent::TLS), gains the ability to do TLS/SSL.

   Time::HiRes
       This module is part of perl since release 5.008. It will be used
       when the chosen event library does not come with a timing source of
       its own. The pure-perl event loop (AnyEvent::Loop) will additionally
       load it to try to use a monotonic clock for timing stability.

   AnyEvent::AIO (and IO::AIO)
       The default implementation of AnyEvent::IO is to do I/O
       synchronously, stopping programs while they access the disk, which
       is fine for a lot of programs.

       Installing AnyEvent::AIO (and its IO::AIO dependency) makes it
       switch to a true asynchronous implementation, so event processing
       can continue even while waiting for disk I/O.

FORK
   Most event libraries are not fork-safe. The ones who are usually are
   because they rely on inefficient but fork-safe "select" or "poll" calls
   - higher performance APIs such as BSD's kqueue or the dreaded Linux
   epoll are usually badly thought-out hacks that are incompatible with
   fork in one way or another. Only EV is fully fork-aware and ensures that
   you continue event-processing in both parent and child (or both, if you
   know what you are doing).

   This means that, in general, you cannot fork and do event processing in
   the child if the event library was initialised before the fork (which
   usually happens when the first AnyEvent watcher is created, or the
   library is loaded).

   If you have to fork, you must either do so *before* creating your first
   watcher OR you must not use AnyEvent at all in the child OR you must do
   something completely out of the scope of AnyEvent (see below).

   The problem of doing event processing in the parent *and* the child is
   much more complicated: even for backends that *are* fork-aware or
   fork-safe, their behaviour is not usually what you want: fork clones all
   watchers, that means all timers, I/O watchers etc. are active in both
   parent and child, which is almost never what you want. Using "exec" to
   start worker children from some kind of manage prrocess is usually
   preferred, because it is much easier and cleaner, at the expense of
   having to have another binary.

   In addition to logical problems with fork, there are also implementation
   problems. For example, on POSIX systems, you cannot fork at all in Perl
   code if a thread (I am talking of pthreads here) was ever created in the
   process, and this is just the tip of the iceberg. In general, using fork
   from Perl is difficult, and attempting to use fork without an exec to
   implement some kind of parallel processing is almost certainly doomed.

   To safely fork and exec, you should use a module such as Proc::FastSpawn
   that let's you safely fork and exec new processes.

   If you want to do multiprocessing using processes, you can look at the
   AnyEvent::Fork module (and some related modules such as
   AnyEvent::Fork::RPC, AnyEvent::Fork::Pool and AnyEvent::Fork::Remote).
   This module allows you to safely create subprocesses without any
   limitations - you can use X11 toolkits or AnyEvent in the children
   created by AnyEvent::Fork safely and without any special precautions.

SECURITY CONSIDERATIONS
   AnyEvent can be forced to load any event model via
   $ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used
   to execute arbitrary code or directly gain access, it can easily be used
   to make the program hang or malfunction in subtle ways, as AnyEvent
   watchers will not be active when the program uses a different event
   model than specified in the variable.

   You can make AnyEvent completely ignore this variable by deleting it
   before the first watcher gets created, e.g. with a "BEGIN" block:

      BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} }

      use AnyEvent;

   Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can
   be used to probe what backend is used and gain other information (which
   is probably even less useful to an attacker than PERL_ANYEVENT_MODEL),
   and $ENV{PERL_ANYEVENT_STRICT}.

   Note that AnyEvent will remove *all* environment variables starting with
   "PERL_ANYEVENT_" from %ENV when it is loaded while taint mode is
   enabled.

BUGS
   Perl 5.8 has numerous memleaks that sometimes hit this module and are
   hard to work around. If you suffer from memleaks, first upgrade to Perl
   5.10 and check wether the leaks still show up. (Perl 5.10.0 has other
   annoying memleaks, such as leaking on "map" and "grep" but it is usually
   not as pronounced).

SEE ALSO
   Tutorial/Introduction: AnyEvent::Intro.

   FAQ: AnyEvent::FAQ.

   Utility functions: AnyEvent::Util (misc. grab-bag), AnyEvent::Log
   (simply logging).

   Development/Debugging: AnyEvent::Strict (stricter checking),
   AnyEvent::Debug (interactive shell, watcher tracing).

   Supported event modules: AnyEvent::Loop, EV, EV::Glib, Glib::EV, Event,
   Glib::Event, Glib, Tk, Event::Lib, Qt, POE, FLTK, Cocoa::EventLoop, UV.

   Implementations: AnyEvent::Impl::EV, AnyEvent::Impl::Event,
   AnyEvent::Impl::Glib, AnyEvent::Impl::Tk, AnyEvent::Impl::Perl,
   AnyEvent::Impl::EventLib, AnyEvent::Impl::Qt, AnyEvent::Impl::POE,
   AnyEvent::Impl::IOAsync, AnyEvent::Impl::Irssi, AnyEvent::Impl::FLTK,
   AnyEvent::Impl::Cocoa, AnyEvent::Impl::UV.

   Non-blocking handles, pipes, stream sockets, TCP clients and servers:
   AnyEvent::Handle, AnyEvent::Socket, AnyEvent::TLS.

   Asynchronous File I/O: AnyEvent::IO.

   Asynchronous DNS: AnyEvent::DNS.

   Thread support: Coro, Coro::AnyEvent, Coro::EV, Coro::Event.

   Nontrivial usage examples: AnyEvent::GPSD, AnyEvent::IRC,
   AnyEvent::HTTP.

AUTHOR
      Marc Lehmann <[email protected]>
      http://anyevent.schmorp.de

You are viewing proxied material from ftp.icm.edu.pl. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.